Exam Engine

Assessment creator that turns a topic and a spec into a graded, printable question paper in 10 to 30 seconds. Teachers pick subject, difficulty, counts per question type; the engine generates, grades, and exports PDF.

Live: https://vedaai-7vks.onrender.com

Stack

Next.js 14 (App Router) · TypeScript · Express · MongoDB (Mongoose) · Redis + BullMQ · Socket.io · OpenAI GPT-4o (JSON mode) · @react-pdf/renderer · Docker Compose

What's interesting here

• Job-queue architecture with live progress. Generation is offloaded to a BullMQ worker so the request returns in milliseconds. A Socket.io room tied to the job ID streams progress events back to the client while questions are produced.
• Strict JSON-mode LLM calls. Every generation request hits GPT-4o with response_format: { type: "json_object" } and a schema contract, so the worker can deserialize directly into Mongoose documents without a brittle regex parser.
• Client-side PDF. Graded papers render with @react-pdf/renderer in the browser, so exports never hit the server and load scales with users, not CPU.

Run locally

# 1. Clone and install
git clone https://github.com/kaniikaaaa/exam-engine.git
cd exam-engine
cp server/.env.example server/.env   # fill OPENAI_API_KEY, MONGO_URI, REDIS_URL

# 2. Start MongoDB + Redis via Docker
docker compose up -d

# 3. Install deps (root runs both client + server via concurrently)
npm install
cd client && npm install && cd ..
cd server && npm install && cd ..

# 4. Run in dev
npm run dev
# client: http://localhost:3000
# server: http://localhost:5000

Project layout

client/   Next.js 14 app, Zustand stores, @react-pdf renderer
server/   Express API, Mongoose models, BullMQ worker, Socket.io
docker-compose.yml   MongoDB + Redis

Status

Live and deployed. Reached final interview round with VedaAI on this project.

License

MIT